CVector 
Revised 12 Dec 2008, for sourceforge release, Larry Andrews and Herbert J. Bernstein
8 Jan 2009 Release 1.0 LCA and HJB
11 Jan 2009 Release 1.0.1 LCA and HJB
21 March 2009 Release 2.0 LCA and HJB
30 May 2009 Release 2.1 LCA and HJB
4 June 2009 Release 2.1.1 LCA and HJB
7 June 2009 Release 2.1.2 LCA and HJB
7 July 2009 Release 2.1.3 LCA and HJB
29 November 2009 Release 2.1.4 LCA
23 April 2010 Release 2.1.5 LCA and HJB
18 July 2010 Release 2.2 HJB
25 July 2010 Release 2.2.1 HJB
31 August 2010 Release 2.3 LCA
7 September 2010 Release 2.3.1 LCA
30 October 2010 Release 2.3.2 LCA
22 March 2011 Release 3.0 LCA and HJB
5 April 2011 Release 3.0.1 LCA and HJB
19 April 2011 Release 3.0.2 HJB
23 April 2011 Release 3.1 HJB
27 September 2011 Release 3.1.1 HJB
LGPL NOTICESThis library is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 2.1 of the License, or (at your option) any later version. This library is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU* Lesser General Public License for more details. You should have received a copy of the GNU Lesser General Public License along with this library; if not, write to the Free Software Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 021101301 USA 
This is a release of an API for finding nearest neighbors among points in spaces of arbitrary dimensions. This release provides a C++ template, TNear.h, and a C library, CNearTree.c, with example/test programs.
Release 3.1.1 adjusted the libtool version from 5:0:1 to 6:0:1 to avoid confusion on the SONAME of the library as requested by Teemu Ikonen for use as a Debian package.
Release 3.1 adjusted the randomization to be based on the depth rather than the population and added an optional detailed height calculation.
Release 3.0.2 added to randomization on insertion when the tree is not well balanced.
Release 3.0.1 updated the diameter calculation and fixed some documentation errors.
Release 3.0 (formerly named Release 2.4) is a major change to NearTree, restructuring the default search from leftfirst to balanced and adding hooks to collect information about the tree.
Release 2.3.2 adds optional returns of vectors of ordinals of found objects
Release 2.3.1 adds Centroid method for Lloyd clustering.
Release 2.3 added methods for clustering.
Release 2.2.1 was a minor revision to Release 2.2 to add an include of limits.h to TNear.h, primarily for MINGW use.
Release 2.2 added support for C code for fixed length string searches using a hamming distance norm, and for spherical and hemispherical geodesic norm based searches. Because of the addition of new type and norm flags, the version 2.2 shared libraries cannot be used to support binaries compiled against earlier headers and viceversa.
Release 2.1.5 was a cleanup update to the 2.1 release of 30 May 2009 to increase portability, in five stages (2.1.1 on 4 June 2009, 2.1.2 on 7 June 2009, 2.1.3 on 7 July 2009, 2.1.4 on 29 November 2009 and 2.1.5 on 23 April 2010) dealing with the following issues:
The 2.1 release was a minor update to the 2.0 release of 21 March 2009 to deal with the following issues:
Release 2.0 was a major update to the 1.0 release of 8 January 2009 to deal with the following issues:
Our thanks to Nicolas Brodu for suggesting the more general handling of the distance type.
Note: As Nicolas Brodu has noted, CNearTree is particularly wellsuited to multithreaded applications. However, if the same CNearTree is to be searched in multiple threads, it is important to complete all insertions and/or delayed insertions before parallel execution of parallel searches.
The NearTree package is available at www.sourceforge.net/projects/neartree. A source tarball is available at downloads.sourceforge.net/neartree/NearTree3.1.1.tar.gz. Later tarballs may be available.
If you decide to simply use the TNear.h header to add nearest neighbor support to C++ code under Visual Studio, be sure to also use the rhrand.h and triple.h headers. It is no longer necessary to define USE_LOCAL_HEADERS, which is automatically defined if _MSC_VER is defined. For unix or MINGW, you will need to use the Makefile and to have libtool on your system. Be warned that the default libtool under Mac OS X will not work for this installation.
When the source tarball is downloaded and unpacked, you should have a directory NearTree3.1.1. To see the current settings for a build execute
make
which should give the following information:
PLEASE READ README_NearTree.txt and lgpl.txt Before making the NearTree libraries and example programs, check that the chosen settings are correct The current C++ and C compile commands are: libtool mode=compile g++ g O2 Wall ansi pedantic \ DCNEARTREE_SAFE_TRIANG=1 I. c libtool mode=compile gcc g O2 Wall ansi pedantic \ DCNEARTREE_SAFE_TRIANG=1 I. c The C API, CNearTree.c, depends on the sourceforge project CVector You are currently setup to use the system defaults for CVector If that is not correct, define the variable CVECTOR_INCLUDE The current library link command is: libtool mode=link gcc versioninfo 6:0:1 \ noundefined rpath /usr/local/lib The current C++ and C library local, and C dynamic and static build commands are: libtool mode=link g++ noundefined g O2 Wall ansi pedantic \ DCNEARTREE_SAFE_TRIANG=1 I. libtool mode=link gcc g O2 Wall ansi pedantic \ DCNEARTREE_SAFE_TRIANG=1 I. libtool mode=link gcc noundefined g O2 Wall ansi pedantic \ DCNEARTREE_SAFE_TRIANG=1 shared I/usr/local/include libtool mode=link gcc g O2 Wall ansi pedantic \ DCNEARTREE_SAFE_TRIANG=1 staticlibtoollibs I/usr/local/include Before installing the NearTree library and example programs, check that the install directory and install commands are correct: The current values are : /usr/local libtool mode=install cp To compile the NearTree library and example programs type: make clean make all To run a set of tests type: make tests To clean up the directories type: make clean To install the library and headers type: make install
If these settings need to be changed, edit Makefile. On some systems, e.g. Mac OS X, the default libtool is not appropriate. In that case you should install a recent version of libtool. The CVector kit has been tested with libtool versions 1.3.5 and 1.5.4. For MINGW, libtool version 2.2.6 and gcc version 4 are needed to work with shared libraries (DLLs). If the system libtool is not to be used, define the variable LIBTOOL to be the path to the libtool executable, e.g. in bash
export LIBTOOL=$HOME/bin/libtool
or in the Makefile
LIBTOOL = $(HOME)/bin/libtool
If you need to include local header files using #include "..." instead of #include <...>, define the variable USE_LOCAL_HEADERS. USE_LOCAL_HEADERS is the default for Visual Studio under Microsoft Windows.
Optionally, you may also define CNEARTREE_FORCEFLIP to maximize tree reorganization on insertion, CNEARTREE_NOFLIP to suppress tree reorganization on insertion, CNEARTREE_NODEFER to make all insertions immediate, CNEARTREE_FORCEPREPUNE to do searches first with a tighter estimate on the search radius, and CNEARTREE_NOPREPRUNE to suppress that behavior. The defaults are to do tree reorganization on insertion, to defer insertions, but not to preprune the search radius.
If you define CNEARTREE_INSTRUMENTED, code will be enabled to track node visits in searching the tree.
The triangle inequality that must be evaluated in building trees and retrieving data may not be evaluated correctly if the range of the three values is extremely large (>10**15 or so for doubles) or may be evaluated differently by some compilers in different parts of a program (due to differing usage of registers). The default in this API is to do the triangle inequality three different ways under the control of CNEARTREE_SAFE_TRIANG
#ifdef CNEARTREE_SAFE_TRIANG #define TRIANG(a,b,c) ( (((b)+(c))(a) >= 0) \  ((b)((a)(c)) >= 0) \  ((c)((a)(b)) >= 0)) #else #define TRIANG(a,b,c) ( (((b)+(c))(a) >= 0)) #endif
Problems with the unsafe definition of TRIANG have been seen in Linux under gcc version 4 and in MS Window under VS 2003. There is a slight performance hit from the triple test. If maximal speed is critical and misidentification of nearest points by relative distance errors of about 1 part in 10**15 is not a serious problem, the definition of DCNEARTREE_SAFE_TRIANG=1 can be removed from the definition of CFLAGS in the Makefile.
NOTE: A range of 10**15 is comparable to the diameter of the earth vs. the separation of two bonded atoms.
This is a revised release of
template <typename T, typename DistanceType=double, int distMinValue=1 > class CNearTree;
implementing the Nearest Neighbor algorithm after Kalantari and McDonald, (IEEE Transactions on Software Engineering, v. SE9, pp. 631634,1983) modified to use recursion for insertions and recursion (original version) or a stack (current version) for searches instead of a doublelinked tree and simplified. The default search algorithm no longer favors the left branch first, but follows the more balanced Kalantari and McDonald approach. The prior search algorithm is available in "Left" versions of the search routines doing a bit less checking for things like is the distance to the right less than the distance to the left.
This template is used to contain a collection of objects. After the collection has been loaded into this structure, it can be quickly queried for which object is "closest" to some probe object of the same type. The major restriction on applicability of the neartree is that the algorithm only works if the objects obey the triangle inequality. The triangle rule states that the length of any side of a triangle cannot exceed the sum of the lengths of the other two sides.
CNearTree is the root class for the neartree. The actual data of the tree is stored in NearTreeNode objects descending from a CNearTree.
The types of objects that can be stored in the tree is quite broad. The biggest limitation is that the objects must reside in some sort of metric space and must obey the triangle rule. They must also be all of the same size because they are stored in an std::vector. If your application requires objects of varying storage, then your best way to use this code is to store pointers or handles and to write your own distance functions. Note that std::string is a pointer type variable and so can be stored directly.
The type of the objects to be stored is the only required template argument. The type of the distance measure (DistanceType) defaults to double. If your application is for an integer type, then the type for DistanceType can be your integer type. This has the potential for speeding the calculations by avoiding FP computation. Other general types can be used if desired, but you may need to also input a value of distMinValue.
The template argument distMinValue must be something that your class will understand as a negative number. The default input is negative one. Internally, that is cast to DistanceType. Since most uses will be for DistanceType to be double, that is a simple conversion. Obviously, for integer types, there is no problem either. The need for this value is to have something internally that is recognizable as smaller than the smallest "distance" that can exist between any two objects in your type. For most users, there is no need to input anything other than the default, 1. 1 must be castable to DistanceType. It seems unlikely that anyone would actually need this optional parameter, but it is here for completeness.
It is a design decision that this class cannot work for unsigned types. Verifying the triangle rule for unsigned types is more complex. Sorry, unsigned types are left as an exercise for the reader.
The user of this class needs to provide at least the following functionality for the template to work. For the builtin numerics of C++, they are provided by the system.
DistanceType Norm( void );  
// a function "Norm( void )" of the templated class // to return DistanceType (usually will return a // "length" of type double) 

operator ( );  // geometrical (vector) difference of two objects 
// a copy constructor would be nice  
// a constructor would be nice  
// a destructor would be nice 
The provided interface is:
#include <TNear.h> CNearTree( void ) // constructor instantiated by something like: CNearTree <T> vTree; for some type T CNearTree( const ContainerType<T> & o) // constructor from containers, e.g. ... CNearTree( const std::vector<T> & o ) // constructor CNearTree( const std::list<T> & o ) // constructor CNearTree( const std::set<T> & o ) // constructor CNearTree( const CNearTree<T> & o ) // constructor CNearTree( const ContainerType<T> & o1, const ContainerType<T> & o2) // constructor merging 2 containers, e.g. ... ~CNearTree( void ) // destructor void clear( void ) // clear the NearTree long GetFlags( void ) const // Get all execution flags void SetFlags( const long flags ) // Set all execution flags long GetFlags( const long mask ) const // Get execution flags within mask void SetFlags( const long flags, const long mask ) // Set execution flags within mask // The available execution flags are static const long NTF_NoPrePrune = 1; //flag to suppress all search prepruning static const long NTF_ForcePrePrune = 2; //flag to force search prepruning static const long NTF_NoFlip = 4; //flag to suppress flips on insert static const long NTF_ForceFlip = 8; //flag to force flips on insert static const long NTF_NoDefer =16; //flag to prevent deferred insert template<typename InputContainer> CNearTree& operator=( const InputContainer& o ) // put container's contents into a NearTree, // wiping out the current contents template<typename InputContainer> CNearTree& operator=( InputContainer& o ) // put container's contents into a NearTree, // wiping out the current contents template<typename InputContainer> CNearTree& operator+=( const InputContainer& o ) // add a container's contents to a NearTree template<typename InputContainer> CNearTree& operator=( const InputContainer& o ) // remove a container's contents from a NearTree template<typename InputContainer> CNearTree& set_symmetric_difference( const InputContainer&, o ) // remove the part of a container's // contents from a NearTree that is // already in the Neartree and add // in the contents of the container // that is not already in the Neartree // i.e. the exclusive or void insert( const T& t ) where t is an object of the type T all inserts are delayed until a search is performed or until an explicit call to CompleteDelayedInsertions is called or a search is called. The purpose is to distribute the objects a bit more randomly. Excessively ordered objects leads to less than optimal trees. Starting with the 2.1 release, places objects in a queue for insertion later when CompleteDelayInsert is called. In earlier releases the default was immediate insertion. The following additional convenience insert template allow insertion of containers of objects template< typename InputContainer > void insert( ContainerType & o ) // e. g. ... void insert( const std::vector<T> & o ) void insert( const std::list<T> & o ) void insert( const std::set<T> & o ) void insert( const CNearTree<T> & o ) iterator NearestNeighbor ( const DistanceType & dRadius, const T& t ) const returns an iterator to the nearest point to the probe point t or end() if there is none bool NearestNeighbor ( const DistanceType& dRadius, T& tClosest, const T& t ) const dRadius is the largest radius within which to search; make it very large if you want to include every point that was loaded. tClosest is returned as the object that was found closest to the probe point (if any were within radius dRadius of the probe) t is the probe point, used to search in the group of points insert'ed return value is true if some object was found within the search radius, false otherwise. If false is returned, tClosest is invalid (at best). iterator FarthestNeighbor ( T& const T& t ) const returns an iterator to the nearest point to the probe point t or end() if there is none bool FarthestNeighbor ( T& tFarthest, const T& t ) const tFarthest is returned as the object that was found farthest from the probe point t is the probe point, used to search in the group of points Insert'ed return value is true if some object was found, false otherwise If false is returned, tFarthest is invalid (at best). iterator LeftNearestNeighbor ( const DistanceType & dRadius, const T& t ) const returns an iterator to the nearest point to the probe point t or end() if there is none bool LeftNearestNeighbor ( const DistanceType& dRadius, T& tClosest, const T& t ) const dRadius is the largest radius within which to search; make it very large if you want to include every point that was loaded. tClosest is returned as the object that was found closest to the probe point (if any were within radius dRadius of the probe) t is the probe point, used to search in the group of points insert'ed return value is true if some object was found within the search radius, false otherwise. If false is returned, tClosest is invalid (at best). iterator LeftFarthestNeighbor ( T& const T& t ) const returns an iterator to the nearest point to the probe point t or end() if there is none bool LeftFarthestNeighbor ( T& tFarthest, const T& t ) const tFarthest is returned as the object that was found farthest from the probe point t is the probe point, used to search in the group of points Insert'ed return value is true if some object was found, false otherwise If false is returned, tFarthest is invalid (at best). The "Left..." versions of NearestNeighbor and FarthestNeighbor are deprecated versions provided for compatibility with earlier releases of NearTree. There are also "Short..." and "LeftShort..." versions of NearestNeighbor to support experimental prepruning logic. The following functions (BelongsToPoints, SeparateByRadius, FindInSphere, FindOutSphere, and FindInAnnulus) all return a container (ContainerType) that can be any standard library container (such as std::vector< T >) or CNearTree. In each case option arguments allow a parallel vector of indices to be returned for each container, giving the indices of the returned objects within the original NearTree. template<typename ContainerType> void BelongsToPoints ( const T& t1, const T& t2, ContainerType& group1, ContainerType& group2 ) template<typename ContainerType> void BelongsToPoints ( const T& t1, const T& t2, ContainerType& group1, ContainerType& group2, std::vector<size_t>& group1_ordinals, std::vector<size_t>& group2_ordinals) returns the points closer to t1 than to t2 in group1 and the rest in group 2 if group1_ordinals and group2_ordinals are provided the ordinals of the found objects in the object store are put into those vectors. The ordinals can be used as indices into the CNearTree itself. template<typename ContainerTypeInside, typename ContainerTypeOutside> void SeparateByRadius ( const DistanceType radius, const T& probe, ContainerTypeInside& inside, ContainerTypeOutside& outside ) template<typename ContainerTypeInside, typename ContainerTypeOutside> void SeparateByRadius ( const DistanceType radius, const T& probe, ContainerTypeInside& inside, ContainerTypeOutside& outside, std::vector<size_t>& inside_ordinals, std::vector<size_t>& outside_ordinals) return the points within radius of the probe in inside and the rest in outside if inside_ordinals and outside_ordinals are provided the ordinals of the found objects in the object store are put into those vectors. The ordinals can be used as indices into the CNearTree itself. long FindInSphere ( const DistanceType& dRadius, ContainerType& tInside, const T& t ) const long FindInSphere ( const DistanceType& dRadius, ContainerType& tInside, std::vector<size_t>& tIndices, const T& t ) const long FindInSphere ( const DistanceType& dRadius, CNearTree< T >& tInside, const T& t ) const long LeftFindInSphere ( const DistanceType& dRadius, ContainerType& tInside, const T& t ) const long LeftFindInSphere ( const DistanceType& dRadius, ContainerType& tInside, std::vector<size_t>& tIndices, const T& t ) const long LeftFindInSphere ( const DistanceType& dRadius, CNearTree< T >& tInside, const T& t ) const dRadius is the radius within which to search; make it very large if you want to include every point that was loaded; tInside is returned as the NearTree or container of objects that were found within a radius dRadius of the probe point if the tIndices argument is given it will be returned as a vector of indices in the near tree of the objects returned. t is the probe point, used to search in the group of points Insert'ed return value is the count of the number of points found within the search radius the "Left..." versions are deprecated versions provided for compatibility with earlier NearTree releases. long FindOutSphere ( const DistanceType& dRadius, ContainerType& tOutside, const T& t ) const long FindOutSphere ( const DistanceType& dRadius, ContainerType& tOutside, std::vector<size_t>& tIndices, const T& t ) const long FindOutSphere ( const DistanceType& dRadius, CNearTree< T >& tOutside, const T& t ) const long LeftFindOutSphere ( const DistanceType& dRadius, ContainerType& tOutside, const T& t ) const long LeftFindOutSphere ( const DistanceType& dRadius, ContainerType& tOutside, std::vector<size_t>& tIndices, const T& t ) const long LeftFindOutSphere ( const DistanceType& dRadius, CNearTree< T >& tOutside, const T& t ) const dRadius is the radius outside of which to search tOutside is returned as the NearTree or container of objects that were found at or outside of radius dRadius of the probe point if the tIndices argument is given it will be returned as a vector of indices in the near tree of the objects returned. t is the probe point, used to search in the group of points Insert'ed return value is the count of the number of points found outside the search radius the "Left..." versions are deprecated versions provided for compatibility with earlier NearTree releases. long FindInAnnulus ( const DistanceType& dRadius1, const DistanceType& dRadius2, ContainerType& tInRing, const T& t ) const long FindInAnnulus ( const DistanceType& dRadius1, const DistanceType& dRadius2, ContainerType& tInRing, std::vector<size_t>& tIndices, const T& t ) const long FindInAnnulus ( const DistanceType& dRadius1, const DistanceType& dRadius2, CNearTree< T >& tInRing, const T& t ) const long LeftFindInAnnulus ( const DistanceType& dRadius1, const DistanceType& dRadius2, ContainerType& tInRing, const T& t ) const long LeftFindInAnnulus ( const DistanceType& dRadius1, const DistanceType& dRadius2, ContainerType& tInRing, std::vector<size_t>& tIndices, const T& t ) const long LeftFindInAnnulus ( const DistanceType& dRadius1, const DistanceType& dRadius2, CNearTree< T >& tInRing, const T& t ) const dRadius1 and dRadius2 are the two radii between which to find data points tInRing is returned as the NearTree or container of objects that were found at or outside of a radius dRadius1 and at or inside of radius dRadius2 of the probe point if the tIndices argument is given it will be returned as a vector of indices in the near tree of the objects returned. t is the probe point, used to search in the group of points Insert'ed return value is the count of the number of points found within the annulus the "Left..." versions are deprecated versions provided for compatibility with earlier NearTree releases. long FindK_NearestNeighbors ( const size_t k, const DistanceType& dRadius, ContainerType& tClosest, const T& t ) long FindK_NearestNeighbors ( const size_t k, const DistanceType& dRadius, ContainerType& tClosest, std::vector<size_t>& tIndices, const T& t ) long FindK_NearestNeighbors ( const size_t k, const DistanceType& dRadius, CNearTree< T >& tClosest, const T& t ) long LeftFindK_NearestNeighbors ( const size_t k, const DistanceType& dRadius, ContainerType& tClosest, const T& t ) long LeftFindK_NearestNeighbors ( const size_t k, const DistanceType& dRadius, ContainerType& tClosest, std::vector<size_t>& tIndices, const T& t ) long LeftFindK_NearestNeighbors ( const size_t k, const DistanceType& dRadius, CNearTree< T >& tClosest, const T& t ) k is the maximum number of nearest neighbors to return. Finds this many if possible dRadius within a sphere defined by dRadius, to search for the knearestneighbors tClosest is returned as the ContainerType or NearTree of the objects found if the tIndices argument is given it will be returned as a vector of indices in the near tree of the objects returned. t is the probe point, used to search in the group of points insert'ed return value is the count of the number of points found within the sphere the "Left..." versions are deprecated versions provided for compatibility with earlier NearTree releases. long FindK_FarthestNeighbors ( const size_t k, ContainerType& tFarthest, const T& t ) long FindK_FarthestNeighbors ( const size_t k, ContainerType& tFarthest, std::vector<size_t>& tIndices, const T& t ) long FindK_FarthestNeighbors ( const size_t k, CNearTree< T >& tFarthest, const T& t ) long LeftFindK_FarthestNeighbors ( const size_t k, ContainerType& tFarthest, const T& t ) long LeftFindK_FarthestNeighbors ( const size_t k, ContainerType& tFarthest, std::vector<size_t>& tIndices, const T& t ) long LeftFindK_FarthestNeighbors ( const size_t k, CNearTree< T >& tFarthest, const T& t ) k is the maximum number of farthest neighbors to return. Finds this many if possible tFarthest is returned as the ContainerType or NearTree of the objects found if the tIndices argument is given it will be returned as a vector of indices in the near tree of the objects returned. t is the probe point, used to search in the group of points insert'ed return value is the count of the number of points found within the sphere the "Left..." versions are deprecated versions provided for compatibility with earlier NearTree releases.
Access Functions: T at ( const size_t n ) const returns the n'th item of the internal data store. This is not guaranteed to be in the order of insertion. T operator[] ( const size_t n ) returns the n'th item of the internal data store. This is not guaranteed to be in the order of insertion. template<typename ContainerType> operator ContainerType ( void ) const returns all of the inserted objects in the tree in a container of type ContainerType. ContainerType can be std::vector<T>, etc, or other containers, including CNearTree<T>. The returned vector contents are not guaranteed to be returned in the order loaded. iterator begin ( void ) const returns an iterator to the beginning of the internal data store iterator end ( void ) const returns an iterator to the end of the data store (one beyond the last item) iterator back ( void ) const returns an iterator to the last data item of the internal data store
Information and special operation functions: void ImmediateInsert( const T&aamp; t ) insert places objects in a queue for insertion later when CompleteDelayInsert is called or a search is called. ImmediateInsert inserts the data immediately into the tree (with the potential of a less balanced tree). ImmediateInsert is not intended for the ordinary user. void CompleteDelayedInsert ( void ) completes insertion for all delayed objects. sqrt(n) of them are inserted by random choice. The rest are inserted in linear order as originally queued. CompleteDelayedInsert is invoked at the beginning of all searches, so the average user will never need to call it. size_t GetDeferredSize ( void ) returns the number of delayed objects that have not yet completed insertion. This is mainly for information about details of the tree. size_t GetTotalSize ( void ) returns the number of objects that have been insert'ed plus those DelayInsert'ed size_t size ( void ) identical to GetTotalSize size_t GetDepth ( void ) returns the maximum tree layers from the root. This is mainly for information about details of the tree. double GetDimEstimate ( void ) // returns an estimate of the Hausdorff dimension double GetDimEstimate ( const double DimEstimateEsd ) // returns an estimate of the Hausdorff dimension // to within the given esd double GetDimEstimateEsd ( void ) // returns an estimate of the esd double GetDiamEstimate ( void ) // returns an estimate of the diameter DistanceType GetMeanSpacing ( void ) // returns an estimate object spacing DistanceType GetVarSpacing ( void ) // returns an estimate object spacing variance size_t GetNodeVisits ( void ) // returns the number of node visits if // CNEARTREE_INSTRUMENTED as defined, 0 otherwise void SetNodeVisits,/b> ( const size_t visits) // set the number of node visits T Centroid ( void ) returns the centroid of a neartree. bool empty ( void ) returns true if the tree is empty, otherwise false
Iterators: Random access iterators are provided for accessing the data in a CNearTree. The most important expected use is to retrieve the objects returned from one of the sphere search functions that can return a CNearTree. However, they can be used with any CNearTree. They should function in a fashion essentially the same as STL iterators. There is no assurance that data will be returned in the order it was loaded, just that it is accessible. This is the list of iterators. The same set is available for const_iterator. iterator ( void ) { }; // constructor iterator ( const const_iterator& s ) iterator& operator= ( const iterator& s ) iterator& operator= ( const const_iterator& s ) iterator operator++ ( const int n ) iterator operator ( const int n ) iterator& operator++ ( void ) iterator& operator ( void ) iterator operator+ ( const long n ) const iterator operator ( const long n ) const iterator& operator+= ( const long n ) iterator& operator= ( const long n ) T operator* ( void ) const bool operator== ( const iterator& t ) const bool operator!= ( const iterator& t ) const bool operator== ( const const_iterator& t ) const bool operator!= ( const const_iterator& t ) const bool operator> ( const iterator& t ) const bool operator> ( const const_iterator& t ) const bool operator< ( const iterator& t ) const bool operator< ( const const_iterator& t ) const const T * const operator> ( void ) const
So a complete program is:
#include "TNear.h" #include <cstdio> void main() { CNearTree< double > dT; double dNear; dT.Insert( 1.5 ); if ( dT.NearestNeighbor( 10000.0, dNear, 2.0 )) printf( "%f\n",double(dNear2.0) ); }and it should print 0.5 (that's how for 2.0 is from 1.5). For more examples of the use of TNear.h, see main.cpp and CNearTreeTest.cpp.
#include <CNearTree.h>
double CNearTreeDistsq ( CNearTreeHandle treehandle, void * coord1, void * coord2 );
double CNearTreeDist ( CNearTreeHandle treehandle, void * coord1, void * coord2 );
int CNearTreeSetNorm ( const CNearTreeHandle treehandle, int treenorm );
int CNearTreeNodeCreate ( const CNearTreeHandle treehandle, CNearTreeNodeHandle * treenodehandle )
int CNearTreeCreate ( CNearTreeHandle * treehandle, size_t treedim, int treetype );
int CNearTreeFree ( const CNearTreeHandle treehandle );
int CNearTreeClear ( CNearTreeHandle * treehandle );
int CNearTreeNodeFree ( CNearTreeNodeHandle * treenodehandle );
int CNearTreeInsert( const CNearTreeHandle treehandle, const void * coord, const void * obj );
int CNearTreeImmediateInsert ( const CNearTreeHandle treehandle, const void * coord, const void * obj );
int CNearTreeDelayedInsert ( const CNearTreeHandle treehandle, const void * coord, const void * obj ); /* ***DEPRECATED*** */
int CNearTreeNodeInsert ( const CNearTreeHandle treehandle, CNearTreeNodeHandle treenodehandle, size_t index; size_t * depth );
int CNearTreeNodeInsert_Flip ( const CNearTreeHandle treehandle, CNearTreeNodeHandle treenodehandle, size_t index; size_t * depth );
int CNearTreeNodeReInsert_Flip ( const CNearTreeHandle treehandle, const CNearTreeNodeHandle treenodehandle, const CNearTreeNodeHandle pntn, size_t * depth );
int CNearTreeCompleteDelayedInsert ( const CNearTreeHandle treehandle ) int CNearTreeZeroIfEmpty ( const CNearTreeHandle treehandle );
int CNearTreeGetSize ( const CNearTreeHandle treehandle, size_t * size );
int CNearTreeGetTotalSize ( const CNearTreeHandle treehandle, size_t * size ); /* ***DEPRECATED*** */
size_t CNearTreeSize ( const CNearTreeHandle treehandle);
int CNearTreeGetDeferredSize ( const CNearTreeHandle treehandle, size_t * size );
int CNearTreeGetDelayedSize ( const CNearTreeHandle treehandle, size_t * size ); /* ***DEPRECATED*** */
int CNearTreeGetDepth ( const CNearTreeHandle treehandle, size_t * depth )
int CNearTreeGetFlags ( const CNearTreeHandle treehandle, long * flags, const long mask )
int CNearTreeSetFlags ( const CNearTreeHandle treehandle, const long flags, const long mask )
int CNearTreeGetMeanSpacing ( const CNearTreeHandle treehandle, double * spacing );
int CNearTreeGetVarSpacing ( const CNearTreeHandle treehandle, double * varspacing );
int CNearTreeCount ( const CNearTreeHandle treehandle, size_t * count );
int CNearTreeNodeCount ( const CNearTreeNodeHandle treenodehandle, size_t * count );
#ifdef CNEARTREE_INSTRUMENTED
int CNearTreeGetNodeVisits ( const CNearTreeHandle treehandle, size_t * visits);
int CNearTreeSetNodeVisits ( const CNearTreeHandle treehandle, const size_t visits );
#endif
int CNearTreeGetDiamEstimate ( const CNearTreeHandle treehandle, double * diamest );
int CNearTreeGetDimEstimateEsd ( const CNearTreeHandle treehandle, double * dimestesd );
int CNearTreeGetDimEstimate ( const CNearTreeHandle treehandle, double * dimest, const double DimEstimateEsd );
int CNearTreeNearestNeighbor ( const CNearTreeHandle treehandle, const double dRadius, void * * coordClosest, void * * objClosest, const void * coord );
int CNearLeftTreeNearestNeighbor ( const CNearTreeHandle treehandle, const double dRadius, void * * coordClosest, void * * objClosest, const void * coord ); /* ***DEPRECATED*** */
int CNearTreeFarthestNeighbor ( const CNearTreeHandle treehandle, void * * coordFarthest, void * * objFarthest, const void * coord );
int CNearTreeFindInSphere ( const CNearTreeHandle treehandle, const double dRadius, CVectorHandle coordInside, CVectorHandle objInside, const void * coord, int resetcount );
int CNearTreeFindTreeInSphere ( const CNearTreeHandle treehandle, const double dRadius, CNearTreeHandle foundInside, const void * coord, int resetcount )
int CNearTreeFindOutSphere ( const CNearTreeHandle treehandle, const double dRadius, CVectorHandle coordOutside, CVectorHandle objOutside, const void * coord, int resetcount );
int CNearTreeFindTreeOutSphere ( const CNearTreeHandle treehandle, const double dRadius, CNearTreeHandle foundOutside, const void * coord, int resetcount )
int CNearTreeFindInAnnulus ( const CNearTreeHandle treehandle, const double dRadiusInner, const double dRadiusOuter, CVectorHandle coordInRing, CVectorHandle objInRing, const void * coord, int resetcount );
int CNearTreeFindTreeInAnnulus ( const CNearTreeHandle treehandle, const double dRadiusInner, const double dRadiusOuter, CNearTreeHandle foundInRing, const void * coord, int resetcount )
int CNearTreeFindKNearest ( const CNearTreeHandle treehandle, const size_t k, const double dRadius, CVectorHandle coordClosest, CVectorHandle objClosest, const void * coord, int resetcount );
int CNearTreeFindKTreeNearest ( const CNearTreeHandle treehandle, const size_t k, const double dRadius, CNearTreeHandle foundClosest, const void * coord, int resetcount )
int CNearTreeFindKFarthest ( const CNearTreeHandle treehandle, const size_t k, const double dRadius, CVectorHandle coordFarthest, CVectorHandle objFarthest, const void * coord, int resetcount );
int CNearTreeFindKTreeFarthest ( const CNearTreeHandle treehandle, const size_t k, const double dRadius, CNearTreeHandle foundFarthest, const void * coord, int resetcount )
int CNearTreeNearest ( const CNearTreeHandle treehandle, double * dRadius, void * * coordClosest, void * * objClosest, const void * coord );
int CNearTreeLeftNearest ( const CNearTreeHandle treehandle, double * dRadius, void * * coordClosest, void * * objClosest, const void * coord ); /* ***DEPRECATED*** */
int CNearTreeFindFarthest ( const CNearTreeHandle treehandle, double * dRadius, void * * coordFarthest, void * * objFarthest, const void * coord );
int CNearTreeObjects ( const CNearTreeHandle treehandle, CVectorHandle * vectorhandle );
void * CNearTreeObjectAt ( const CNearTreeHandle treehandle, size_t index );
int CNearTreeCoords ( const CNearTreeHandle treehandle, CVectorHandle * vectorhandle );
void * CNearTreeCoordAt ( const CNearTreeHandle treehandle, size_t index );
The NearTree API works with coordinate vectors in an arbitrary number of dimensions. Each neartree is accessed by a pointer of type CNearTreeHandle which points to a struct of type CNearTree, which points to a tree of nodes of type CNearTreeNode:
typedef struct _CNearTreeNode { size_t m_indexLeft; /* index of left coords in m_CoordStore and of left object in m_ObjectStore */ size_t m_indexRight; /* index of right coords in m_CoordStore and of right object in m_ObjectStore */ double m_dMaxLeft; /* longest distance from the left object to anything below it in the tree */ double m_dMaxRight; /* longest distance from the right object to anything below it in the tree */ struct _CNearTreeNode * m_pLeftBranch; /* tree descending from the left object */ struct _CNearTreeNode * m_pRightBranch; /* tree descending from the right object */ int m_iflags; /* flags */ } CNearTreeNode; typedef CNearTreeNode * CNearTreeNodeHandle; typedef struct { CNearTreeNodeHandle m_ptTree; /* pointer to the actual tree */ size_t m_szdimension; /* dimension of the coordinates */ size_t m_szsize; /* size of this tree */ size_t m_szdepth; /* depth of this tree */ int m_iflags; /* flags */ CVectorHandle m_ObjectStore; /* all inserted objects */ CVectorHandle m_CoordStore; /* all inserted coordinates */ CVectorHandle m_DelayedIndices;/* objects queued for insertion */ } CNearTree; typedef CNearTree FAR * CNearTreeHandle; /* Execution Control Flags */ #define CNTF_NOPREPRUNE 0x10000L /*flag to suppress all search prepruning */ #define CNTF_FORCEPREPRUNE 0x20000L /*flag to force search prepruning */ #define CNTF_NOFLIP 0x40000L /*flag to suppress flips on insert */ #define CNTF_FORCEFLIP 0x80000L /*flag to force flips on insert */ #define CNTF_NODEFER 0x100000L /*flag to prevent deferred insert */The internal operation of the API depends on the function CNearTreeDist that returns the distance (L1, L2 or Linfinity) between two coordinate vectors as a double according to the parameters of the given tree. Note that the tree may store the coordinates as integers or as doubles, but the distance is always computed as a double. If this function is replaced by a user function, it is important that the replacement obey the triangle inequality.
A neartree is created by CNearTreeCreate and freed by CNearTreeFree. treedim is the dimension of the coordinate vectors and treetype is one of the three predefined constants CNEARTREE_TYPE_DOUBLE for double or CNEARTREE_TYPE_INTEGER for integer or CNEARTREE_TYPE_STRING, optionally ORed with CNEARTREE_NORM_L1, CNEARTREE_NORM_L2 or CNEARTREE_NORM_LINF for L1, L2 or Linfinity norms, CNEARTREE_NORM_SPHERE or CNEARTREE_NORM_HSPHERE for a spherical or hemispherical norm (L1norm combination of radial and spherical/hemispherical triangle distances), or CNEARTREE_NORM_HAMMING for the stringHamming distance norm (add one for each differing character position).
Starting with release 2.1, all insertions are delayed by default, unless the insertions is done by a call to CNearTreeImmediateInsert. The insertions that have been queued are completed by a call to CNearTreeCompleteDelayedInsert or by any search. The insertions are actually done in a randomized order, either for an initial block of sqrt(#queue) by default. or for the entire queue if the flag CNEARTREE_DEFER_ALL is ored with treetype.
Starting with release 3 (formerly called release 2.4) optionally, you may also define CNEARTREE_FORCEFLIP to maximize tree reorganization on insertion, CNEARTREE_NOFLIP to suppress tree reorganization on insertion, CNEARTREE_NODEFER to make all insertions immediate, CNEARTREE_FORCEPREPUNE to do searches first with a tighter estimate on the search radius, and CNEARTREE_NOPREPRUNE to suppress that behavior. The defaults are to do tree reorganization on insertion, to defer insertions, but not to preprune the search radius. If you define CNEARTREE_INSTRUMENTED, code will be enabled to track node visits in searching the tree.
The flags CNEARTREE_DEFER_ALL and CNEARTREE_FLIP used in prior releases are deprecated, but are still defined. They have no effect.
When first created, a neartree has no right or left node and with the dMaxbelow set to negative values so that any match found will be stored since it will greater than the negative value. The tree is then populated by calls to CNearTreeInsert, with each call providing a coordinate vector coord and an optional object pointer obj. The API copies the coordinate vector, but does not copy the object. Later, when a search is requested or an explicit call to CNearTreeCompleteDelayedInsert is made, the tree is populated in the order left, right and then the nearer child, working from a randomized selection from the items queued for insertion.
Optionally, the actual insertions may done immediately by calling CNearTreeImmediateInsert instead of CNearTreeInsert. For upwards compatibility of the library for existing code, the deprecated CNearTreeDelayedInsert is provided as an deprecated alternate call to CNearTreeInsert.
The neartree is searched for the nearest or farthest coordinate vector in the neartree to a given probe coordinate vector coord by CNearTreeNearestNeighbor and CNearTreeFarthestNeighbor, respectively. Starting with release 3, the search is balanced, following the left or right branch first depending on which child node is closest. The former leftfirst behavior is deprecated, but still available in CNearLeftTreeNearestNeighbor. The given radius confines the search to a sphere around the probe. If more than a single extremal coordinate point is needed, CNearTreeFindInSphere can be used to obtain a CVector result vector of all the coordinate vectors that satisfy the constraint of being within a specified radius, or CNearTreeFindOutSphere can be used to obtain a CVector result vector of all the coordinates that satisfy the constraint of being outside a specified radius. CNearTreeFindIn Annulus can be used to obtain a CVector result vector of all the coordinates that satisfy the constraint of being between two specified radii from the probe. CNearTreeFindKNearest can be used to obtain a CVector result vector of the k coordinates closest to the probe point such that all results are within the specified radius of the probe point, or CNearTreeFindKFarthest to obtain a CVector result vector of the k coordinates farthest from the probe point such that all results are at or outside the specified radius of the probe point. The vectors themselves are not copied into the result vector. If the parameter resetcount is true (non zero) the result vector is cleared before the search. A CVector result vector of the matching object pointers is returned if objs is not NULL. Aternatively the forms CNearTreeFindTreeInSphere, CNearTreeFindTreeOutSphere, CNearTreeFindTreeInAnnulus, CNearTreeFindKTreeNearest, CNearTreeFindKTreeFarthest can be used to obtain CNearTrees rather than CVectors of results. The functions CNearTreeNearest and CNearTreeFindFarthest implement CNearTreeNearestNeighbor and CNearTreeFarthestNeighbor, respectively, adjusting the radius of the search while the search is in progress and are not normally used by users.
The size of the tree as a count of objects can be obtained using the function NearTreeGetSize or the macro NearTreeSize. The size of the tree as a count of nodes and the depth of the tree can be obtained using the functions CNearTreeCount and CNearTreeGetDepth. Estimates of the Hausdorff dimension, the esd of that estimate, the diameter, the spacing and the variance of the spacing can be obtained with CNearTreeGetDimEstimate, CNearTreeGetDimEstimateEsd, CNearTreeGetDiamEstimate, CNearTreeGetMeanSpacing and CNearTreeGetVarSpacing.
To create a neartree for 3dimensional vectors of doubles:
#include <CNearTree.h> CNearTreeHandle treehandle; int bReturn; ... bReturn = !CNearTreeCreate(&treehandle,3,CNEARTREE_TYPE_DOUBLE);
To insert a copy of a 3dimensional vector of doubles into this tree, with no associated object:
double v[3]; ... v[0] = 1.; v[1] = 2.; v[2] = 3.; bReturn = !CNearTreeInsert(treehandle,&v[0],NULL);
To search for the nearest neighbor to a probe vector vSearch in a radius of 3., returning a pointer to the resulting vector in vBest:
double * vBest; void * vvBest; double vSearch[3]; double dRad = =3.; ... if ( !CNearTreeNearestNeighbor(treehandle,dRad,&vvBest,NULL,vSearch)) { vBest = (double *)vvBest; }
Note the use of a separate void * vvBest instead of a cast of &vBest to avoid compiler type punning warnings.
For more examples of the use of CNearTree.c, see main.c and CNearTreeTest.c in the release kit.
rhrand.h is a portable pseudorandom number generator based one by Rob Harrison, derived from "one in J.M.Hammersley and D.C. Handscomb, 'Monte Carlo Methods,' Methuen & Co., London and Wiley & Sons, New York, 1964, p47". See also, D. E. Knuth "The Art of Computer Programming", Volume 2, "Seminumerical Alogorithms, Third Edition, AddisonWesley, Reading MA, 1997.
rhrand.h is a header file in which a C++ class, RHrand, is defined, and a C struct typedef CRHrand is defined.
The C++ interface is
static const int RHRAND_MAX = 32767; /* the integer range accessible as RHrand::RHRAND_MAX */ RHrand(void) /* the default constructor */ RHrand( const int iseed ) /* a constructor to start with the given seed */ ~RHrand( void) /* a destructor */ void srandom( const int iseed) /* reset the generator based on the given seed */ double urand( void ) /* return a random double uniformly distributed in [0,1) */ int random ( void ) /* return a random integer uniformly distributed in [0, RHRAND_MAX1] */
In C++ code, typical use is
#include <rhhand.h> RHrand rhr; ... x = rhr.urand();
The C interface is suppressed in RHRAND_NOCCODE is defined. Otherwise the C interface is based on defining a struct of type CRHRrand and calling macros that refer to a handle of type RCRHrandHandle.
typedef struct CRHrand_ { /* the struct used in random number generattion */ double buffer[55]; int indx; int jndx; int kndx; double dTemp; } CRHrand; typedef CRHrand * CRHrandHandle; /* the type to be used in maro calls */ #define CRHRAND_MAX 32767 /* the integer range */ #define CRHrandSrandom(randhandle,iseed) ... /* a macro to call to initialize CHRrandHandle randhandle using see int iseed */ #define CRHrandUrand(randhandle) ... /* a macro to return a random double uniformly distributed in [0,1) */ #define CRHrandRandom(randhandle) ((int)(CRHrandUrand(randhandle)*(double)CRHRAND_MAX)) /* a macro to return a random integer uniformly distributed in [0, CRHRAND_MAX1] */
Typical use is
#include <rhhand.h> CRHrand rhr; ... CRHrandSrandom(&rhr, 0 ); ... x = CRHrandUrand(&rhr);